Purpose
To compare the outcomes of Pseudomonas aeruginosa keratitis (PAK) in contact lens wearers (CLWs) and non–contact lens wearers (non-CLWs) and identify risk factors for poor visual acuity (VA) outcomes in each group.
Design
Retrospective cohort study
Methods
Two hundred fourteen consecutive cases of PAK were included between January 2006 and December 2019. Clinical features, microbiologic results, and treatment course were compared between CLW and non-CLW groups. Analyses of clinical features predicting poor final VA were performed.
Results
This study identified 214 infected eyes in 207 patients with PAK, including 163 eyes (76.2%) in CLWs and 51 eyes (23.8%) in non-CLWs. The average age was 39.2 years in CLWs and 71.9 years in non-CLWs ( P < .0001). The average logMAR visual acuity (VA) at presentation was 1.39 in CLWs and 2.17 in non-CLWs ( P < .0001); average final VA was 0.76 in CLWs and 1.82 in non-CLWs ( P < .0001). Stromal necrosis required a procedural or surgical intervention in 13.5% of CLWs and 49.0% of non-CLWs ( P < .0001). A machine learning–based analysis yielded a list of clinical features that most strongly predict a poor VA outcome (worse than 20/40), including worse initial VA, older age, larger size of infiltrate or epithelial defect at presentation, and greater maximal depth of stromal necrosis.
Conclusions
Non-CLWs have significantly worse VA outcomes and required a higher rate of surgical intervention, compared with CLWs. Our study elucidates risk factors for poor visual outcomes in non-CLWs with PAK.
I nfectious keratitis is an important cause of visual disability and blindness worldwide. Pseudomonas aeruginosa (PA) ranks among the most common causative organisms of bacterial keratitis; it is the number 1 organism associated with contact lens (CL) wear. , In a recent study of 3,004 patients with infectious keratitis, PA was the second most common cause of bacterial keratitis, responsible for 18% of cases. Among 558 patients hospitalized with microbial keratitis in Taiwan, Pseudomonas was cultured in 28% of patients and was the most commonly isolated organism.
PA produces the most severe cases of bacterial keratitis , likely because of its extensive arsenal of virulence factors and quorum-sensing regulatory systems. Readily able to modify its pathogenicity, PA has a large genome (>6 Mbp), well-conserved core genome regions, and many accessory genome regions with extensive plasticity. Therefore, there are increasing reports of multidrug-resistant and pandrug-resistant P aeruginosa keratitis (PAK), with longer treatment courses and worse clinical outcomes. ,
Often progressing rapidly, PAK can have severe consequences, including corneal scarring and perforation. PAK isolates have demonstrated adaptations to survive in environmental water, as well as cytotoxic and invasive abilities. , Patients with PAK commonly present with large, central infiltrates associated with severe pain, discharge, epithelial staining, and anterior chamber inflammation , ; the classic finding is a “ring”-shaped, “soupy”-appearing stromal infiltrate of polymorphonuclear leukocytes. Several features may help to differentiate PAK from infectious keratitis caused by other species, including the presence of a hypopyon, rapid stromal thinning, yellow-green purulent discharge, and a ground-glass appearance of the adjacent corneal stroma. , A major risk factor for infectious keratitis in general, CL wear has an especially strong association with PAK. The risk of CL-associated keratitis is amplified by sleeping in CL, smoking, poor handwashing prior to lens insertion, and nonhygienic lens cleaning practices. , , , With increasing rates of CL wear, the incidence of PAK has likewise increased. , Accordingly, many studies have focused on the role of CL wear in the pathogenesis of PAK. , Primary risk factors for infectious keratitis in patients not wearing CL include ocular surface disease, ocular trauma, and prior ocular surgery. ,
Interestingly, although the rate of PAK has increased over time, its severity has decreased. Given the temporal correlation among increasing CL wear, rising incidence, and decreasing severity of PAK, we hypothesized that PA would cause more severe infections in non–contact lens wearers (non-CLWs) compared with contact lens wearers (CLWs). Thus, we asked whether we would see worse visual acuity (VA) outcomes and increased surgical interventions for corneal stromal necrosis in non-CLWs compared with CLWs. To this end, we compared the epidemiologic, clinical, and outcomes profiles for CLWs and non-CLWs with culture-positive PAK. Further, we sought to understand which clinical features would predict worse VA outcomes in CLWs and non-CLWs.
Methods
A retrospective review was performed on the medical records of all patients diagnosed with culture-positive PAK at the University of Pittsburgh Medical Center between January 2006 and September 2019. The study was approved by the Institutional Review Board of the University of Pittsburgh and followed the tenets of the Declaration of Helsinki. Clinical data were collected for each patient, including CL wear status, clinical features, microbiological results, treatment, and outcomes. Demographic features were recorded, including gender and age. Length of time between symptom onset and presentation was recorded, along with use of topical antibiotics and/or steroids before presentation. Ocular data included laterality, location of corneal involvement, size of infiltrate and epithelial defect at presentation, presence and size of hypopyon, presence of stromal thinning, and scleral involvement. Best-corrected Snellen VA was recorded at presentation and after resolution, including any improvement with pinhole, as many patients were CLWs presenting without correction. VA was converted to logMAR using estimates for non-numerical values of CF (counting fingers) = 2.0, HM (hand motions) = 2.3, LP (light perception) = 2.7, and NLP (no light perception) = 3.0.
Microbiological results were recorded, including detected organism(s). Results were recorded for Gram and Giemsa staining of corneal smears on glass slides. All antimicrobial treatments were recorded, including their duration. Susceptibilities of PAK isolates to various antibiotics were recorded, and the presence of any coinfection(s) was determined. Interventions for progressive stromal thinning were documented, including cyanoacrylate glue, amniotic membrane transplantation, full-thickness corneal transplantation for tectonic support, evisceration, or enucleation. Finally, outcomes were documented, including corneal perforation, persistent epithelial defect, and resolution with or without stromal scarring.
Statistical Analysis
Data were stratified into contact lens wearers (CLWs) and non–contact lens wearers (non-CLWs), and clinical variables were compared between the 2 groups. Statistical analysis was conducted in Microsoft Excel 2013 (v15.0.5249.1000), GraphPad Prism (v8.4.3), and Python (v3.7.0) using the packages pandas (v3.6), numpy (v1.18.5), matplotlib (v3.2.2), scipy (v1.4.1), and seaborn (v0.10.1). Machine learning models were created and analyzed using Python with the sklearn (v0.22.2) package. Continuous variables are stated as mean ± standard deviation, and binary variables are stated as number (percentage).
Chi-squared and Student t tests were used to calculate P values for binary and continuous variables, respectively. Bonferroni correction was used to adjust the threshold level of significance; thus, P <.001 was considered statistically significant for the primary analysis. Medical treatment regimens were compared, with P ≥.05 used to establish no difference between the 2 groups.
Univariate odds ratios were calculated for presenting features to investigate their association with a final VA worse than 20/40. To analyze continuous variables, we assigned threshold values to convert them into binary variables. Spearman rank correlation test was used to calculate rho and P values for stratified continuous variables. The association between binary variables and VA outcomes was analyzed separately for CLWs and non-CLWs by determining the mean average of the final logMAR VA for both “true” and “false” values of each binary variable.
Finally, a series of machine learning-based models were iterated to develop a predictive model for final VA worse than 20/40. Several architectures were tested (including Support Vector Machine, K Nearest Neighbour, XGBoost), and a Random Forest classifier validated with 5-fold cross-validation was selected as the best predictive model, which is presented in this manuscript. The Random Forest was initialized from the sklearn library with 100 maximum trees, using the gini impurity criterion, and minimum sample split of 2. From the available 191 cases for training, the data set was split into 80% train, 20% test data. Given the small data set we chose to use 5-fold cross validation to maximize utilization of the given data set, whereas ensuring the model will be capable of generalizing to future unseen data.
Results
Baseline Characteristics and Historical Features
A total of 214 infected eyes in 207 patients with PAK were identified. Among these, 163 infections (76.2%) occurred in CLWs and 51 (23.8%) in non-CLWs. Among the CLWs, 161 eyes (98.8%) wore soft contact lenses and 2 (1.2%) wore rigid gas-permeable lenses. Overall, 200 cases (93.5%) were unilateral, 6 bilateral (5.6%), and 1 sequential (0.9%). Five bilateral cases occurred in CLWs and 1 in a non-CLW. The sole sequential infection occurred first in the left eye and then right eye 14 months later in a patient who intermittently wore a bandage contact lens over a keratoprosthesis in the left eye only.
The demographics and presenting examination findings of patients in this study are summarized in Table 1 , including a comparison of the entire study population, CLWs, and non-CLWs. Initial logMAR VA was 1.57±1.00 in the entire study population. At presentation, CLWs had a logMAR of 1.39±0.98, and non-CLWs had a logMAR of 2.17±0.82 ( P < .0001). Among these patients, 16 CLWs (9.9%) and 18 non-CLWs (37.5%) had light perception or no light perception vision. At presentation, the surface area occupied by the corneal stromal infiltrate was 18.9±27.9 mm 2 , with an epithelial defect size of 15.7±22.2 mm 2 . Table 2 summarizes the historical features and potential risk factors for PAK for the patients in this study.
| Variable | Entire Study Population | Contact Lens Wearers | Non–Contact Lens Wearers | P Value |
|---|---|---|---|---|
| No. of eyes | 214 | 163 | 51 | |
| Age, y | ||||
| Mean ± SD | 47.0 ± 23.3 | 39.2 ± 18.9 | 71.9 ± 17.9 | <.0001 |
| Range | 13-97 | 13-89 | 30-97 | |
| 13-39, n (%) | 93 (43.5) | 90 (55.2) | 3 (5.9) | |
| 40-59, n (%) | 58 (27.1) | 47 (28.8) | 11 (21.6) | |
| 60-79, n (%) | 34 (15.9) | 18 (11.0) | 16 (31.4) | |
| 80-97, n (%) | 29 (13.6) | 8 (4.9) | 21 (41.2) | |
| Sex, n (%) | ||||
| Male | 87 (40.7) | 63 (38.7) | 24 (47.1) | .29 |
| Female | 127 (59.3) | 100 (61.3) | 27 (52.9) | |
| Laterality, n (%) | ||||
| Right eye | 99 (46.3) | 77 (47.2) | 22 (43.1) | .61 |
| Left eye | 115 (53.7) | 86 (52.8) | 29 (56.9) | |
| Time to presentation, d, mean ± SD | 3.25±3.30, | |||
| n=211 | 2.77±2.79, | |||
| n=163 | 4.90±4.26, | |||
| n=48 | <.0001 | |||
| Presenting exam findings, mean ± SD | ||||
| VA a (logMAR) | 1.57 ± 1.00, n=210 | 1.39 ± 0.98, n=162 | 2.17 ± 0.82, n=48 | <.0001 |
| Infiltrate size b (mm 2 ) | 18.9±27.9, n=194 | 15.3±23.7, n=151 | 31.5±36.9, n=43 | .0007 |
| Epithelial defect size b (mm 2 ) | 15.7±22.2, n=201 | 13.6±20.6, n=154 | 22.6±26.0, n=47 | .014 |
| Hypopyon | ||||
| Presence, n (%) | 107 (50.0), n=214 | 80 (49.1), n=163 | 27 (52.9), n=51 | .63 |
| Size, mm, mean ±SD | 1.52±1.31, n=107 | 1.47±1.30, n=80 | 1.67±1.36, n=27 | .50 |
| Location of ulcer | n=204 | n=156 | n=48 | |
| Diffuse, n (%) | 23 (11.3) | 12 (7.7) | 11 (22.9) | .0035 |
| Central, n (%) | 80 (39.2) | 66 (42.3) | 14 (29.2) | .10 |
| Paracentral, n (%) | 63 (30.9) | 50 (32.1) | 13 (27.1) | .51 |
| Peripheral, n (%) | 38 (18.6) | 28 (18.0) | 10 (20.8) | .65 |
a Best-corrected visual acuity including pinhole visual acuity. For non-numerical values, logMAR was estimated as CF (counting fingers) = 2.0, HM (hand motions) = 2.3, LP (light perception) = 2.7, and NLP (no light perception) = 3.0, per Lange and colleagues. 26
b Infiltrate and epithelial defect size are reported as area in square millimeters, which was estimated as a circle with radius equivalent to the average of the recorded dimensions.
| Potential Risk Factors | Entire Study Population | Contact Lens Wearers | Non–Contact Lens Wearers | P Value |
|---|---|---|---|---|
| Number of eyes | 214 | 163 | 51 | |
| Eye drops at presentation | ||||
| Antibiotics, n (%) | 60 (28.0) | 44 (27.0) | 16 (31.4) | .54 |
| Corticosteroids, n (%) | 29 (13.6) | 12 (7.4) | 17 (33.3) | <.0001 |
| Glaucoma drops, n (%) | 16 (7.5) | 3 (1.8) | 13 (25.5) | <.0001 |
| No. of glaucoma drop applications per day, mean±SD | 2.6±1.7, n=16 | 2.3±2.3, n=3 | 2.6±1.7, n=13 | .81 |
| Ocular history, n (%) | ||||
| Recent ocular trauma | 30 (14.0) | 23 (14.1) | 7 (13.7) | .94 |
| Corneal epithelial disease | 41 (19.2) | 14 (8.6) | 27 (52.9) | <.0001 |
| Eyelid laxity | 10 (4.7) | 3 (1.8) | 7 (13.7) | .0004 |
| Prior corneal transplant | 14 (6.5) | 3 (1.8) | 11 (21.6) | <.0001 |
| Prior corneal ulcer | 18 (8.4) | 15 (9.2) | 3 (5.9) | .46 |
| Glaucoma | 16 (7.5) | 3 (1.8) | 13 (25.5) | <.0001 |
| Systemic history, n (%) | ||||
| Herpes infection | 9 (4.2) | 7 (4.3) | 2 (3.9) | .91 |
| Immunosuppression | 37 (17.3) | 19 (11.7) | 18 (35.3) | <.0001 |
| Cancer history | 12 (5.6) | 4 (2.5) | 8 (15.7) | .0003 |
| Autoimmune disease | 27 (12.6) | 15 (9.2) | 12 (23.5) | .0072 |
| Social history, n (%) | ||||
| Institutionalization | 12 (5.6) | 3 (1.8) | 9 (17.6) | <.0001 |
| Medical facility | 8 (3.7) | 0 (0.0) | 8 (15.7) | <.0001 |
| Incarceration | 4 (1.9) | 3 (1.8) | 1 (2.0) | .96 |
| Active smoker | 48 (22.4) | 40 (24.5) | 8 (15.7) | .19 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
